Navitas Semiconductor announced a new DC-DC power delivery board (PDB) based on GaNFast™ technology that enables direct conversion from 800 V to 6 V in a single stage, eliminating the need for a traditional 48 V intermediate bus converter.

The new architecture is designed to support next-generation AI data center platforms, particularly those aligned with NVIDIA infrastructure, where increasing rack power densities demand more efficient and compact power delivery solutions.

By removing the intermediate 48 V conversion stage, the system reduces power losses, improves reliability and frees up valuable board space for compute and memory components. This direct 800 V-to-6 V conversion also improves overall system efficiency and simplifies the power delivery chain for high-performance AI workloads.

The power delivery board is designed to achieve up to 96.5 percent peak efficiency at full load while operating at a switching frequency of 1 MHz. It delivers a power density of approximately 2,100 W/in³ and features an ultra-low profile, enabling close integration with GPU boards to enhance transient performance and power distribution efficiency.

The design uses 16 650 V GaNFast FETs in a dual-cooled DFN8×8 package on the primary side, configured in a stacked full-bridge topology. On the secondary side, center-tapped outputs utilize 25 V silicon MOSFETs. The high switching frequency allows the use of smaller passive components and planar magnetics, contributing to higher power density and compact system design.

The platform aligns with the industry transition toward 800 V DC data center architectures, driven by the increasing power demands of AI and accelerated computing. By enabling direct conversion to low-voltage rails such as 6 V or 12 V, the solution reduces conversion stages and improves end-to-end efficiency compared to conventional architectures.

Navitas stated that this development builds on its earlier 800 V-to-50 V platform and represents a further step toward simplified, high-efficiency power delivery systems for AI infrastructure.

Original – Navitas Semiconductor